RbfCenterInitialization
Enums · AiDotNet.Enums
Where RBFLayer puts its centres before training.
For Beginners
Each RBF unit reacts to inputs close to its centre. Putting the centres on real examples guarantees every unit starts out reacting to something.
How It Works
An RBF network’s centres are points in the input space (Broomhead and Lowe 1988; Moody and Darken 1989). Placed at random near the origin they can sit far from every input, and with Gaussian units that leaves every unit at exp(-large) = 0 for every input. The papers place them on the training data, which is the default here.
Fields
| Field | Summary |
|---|---|
RbfCenterInitialization AsInitialized |
Use the centres as they are: the random initialization, values set through SetParameters or a loaded model, or centres already placed from data. |
RbfCenterInitialization FromFirstBatch |
The first batch the layer sees supplies the centres: evenly spaced rows of that batch, or one centre per row when it has fewer rows than centres (the others keep their random initialization), after which the widths are set from the spread o… |
int value__ |